Explainability as statistical inference

Authors: Hugo Henri Joseph Senetaire, Damien Garreau, Jes Frellsen, Pierre-Alexandre Mattei

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We propose new datasets with ground truth selection which allow for the evaluation of the features importance map and show experimentally that multiple imputation provides more reasonable interpretations.
Researcher Affiliation Academia 1Department of Applied Mathematics and Computer Science, Technical University of Denmark, Denmark 2Universit e Cˆote d Azur, Inria, Maasai, LJAD, CNRS, Nice, France. Correspondence to: Hugo Henri Joseph Senetaire <hhjs@dtu.dk>.
Pseudocode No No pseudocode or clearly labeled algorithm block found.
Open Source Code No No explicit statement or link providing concrete access to the source code for the methodology described in this paper.
Open Datasets Yes For each dataset, we generate 5 different datasets containing each 10,000 train samples and 10, 000 test samples.
Dataset Splits Yes The split between train and validation is split randomly with proportion 80%, 20%. Hence, the train dataset of the switching panels input contain 48,000 images, the validation dataset contains 12,000 images and the test dataset 10,000 images.
Hardware Specification No No specific hardware details (e.g., GPU models, CPU types, memory) used for running experiments are provided.
Software Dependencies No The paper mentions software components like "Adam", "U-Net", "Quickshift", "SHAP", "FASTSHAP", "Sklearn", but does not provide specific version numbers for these or other key software dependencies.
Experiment Setup Yes We trained all the methods for 1000 epochs using Adam for optimisation with a learning rate 10 4 and weight decay 10 3 with a batch size of 1000.